Multi-Modal Module Model Estimation Summary
Prepared by RSG, Inc.
March 8, 2022
Introduction
This document describes the re-estimation of the VE Multimodal Module (VE-MM) using the 2017 National Household Travel Survey (2017 NHTS). The original version of the VE-MM was estimated using the 2009 NHTS.
The models described in this memo include:
- regression models used to predict household average annual daily vehicle miles traveled (AADVMT)
- models used to predict transit, bicycle, and walking trips and person miles traveled (note that these are not included in this draft)
Estimation Data
Data Sources
Data used to estimate the VE-MM models came from several sources. The 2017 NHTS confidential data were obtained from ORNL via FHWA. The data were provided with Census Block Group which allowed data from the EPA Smart Location Database, version 3 (SLD) to be connected to the household. Additional spatial information at the metropolitan area level were added to the data including transit revenue miles from the National Transit Database and freeway lane miles from FHWA published data.
AADVMT and DVMT Data
The AADVMT model is estimated using the total annual miles driven by vehicles owned by the NHTS households divided by 365, instead of estimating a daily VMT model based on the reported travel during the survey day. The distributions of those travel amounts differs in the data.
In particular there are a lot of “zero days” in the survey data for the survey day. There are also more very long days in the survey data with more than 200 miles of travel reported. With the averaging over the year that takes place by using AADVMT, a lot of the day to day variability within a household’s travel is removed. The final chart compares the none-zero day distribution and other than the extremes at the low end and high end of the distribution, the distributions are reasonably comparable.
Figure 1: Distribution of AADVMT per Household in the 2017 NHTS
Figure 2: Distribution of Survey Day DVMT per Household in the 2017 NHTS
Figure 3: Comparison between AADVMT and Survey Day DVMT per Household in the 2017 NHTS
AADVMT Outlier Filtering
The households with the highest 1% of AADVMT were excluded from the estimation dataset. The exclusion of outliers has the effect of reducing the average AADVMT per household. The table below shows the impact on the average AADVMT per household for metro and non-metro households. The charts shows the change in the binned distribution of households for metro and non-metro households.
All of the analysis in the remainder of this document is based on the 2017 NHTS data with outliers removed, consistent with the set of households used for estimation.
| Outlier Status | Metro HH | Non-Metro HH | Metro Avg. AADVMT | Non-Metro Avg. AADVMT |
|---|---|---|---|---|
| Included | 71740 | 56646 | 47.54 | 61.01 |
| Outlier | 617 | 693 | 359.50 | 293.96 |
| Total | 72357 | 57339 | 50.85 | 64.96 |
Figure 4: Impacts of Outlier Filtering on the AADVMT per Household in the 2017 NHTS
Population Density (D1B)
One of the important explanatory variables for household AADVMT is the population density of the neighborhood in which the household resides. The 2017 NHTS data were linked to the SLD and the D1B variable, which contains the Census Block Group population density in units of persons per acre. The ranges of density varies significantly in the sample of NHTS households. As is expected, most non-metro households live in low density areas while the range of densities for metropolitan households is generally higher but covers several orders of magnitude. The following box plot, plotted on a log scale shows the distribution of households by population density.
Figure 5: Distribution of Households in the 2017 NHTS by Population Density
The following charts show the relationship between AADVMT in low, medium, and high density areas, split by metro and non-metro households in low and medium density areas, and showing just metropolitan households in the high density chart. The trends in the charts show a clear relationship between density and AADVMT: as density increases, AADVMT tends to fall. There are some anomalous values, particular for non-metropolitan households, due to small sample sizes.
Figure 6: 2017 NHTS AADVMT by Density for Households in Low Density Neighborhoods
Figure 7: 2017 NHTS AADVMT by Density for Households in Medium Density Neighborhoods
Figure 8: 2017 NHTS AADVMT by Density for Households in High Density Neighborhoods
Household Income
Household income has a positive impact on the amount of travel a household makes, with higher income households traveling more. The first chart shows the distribution of income levels amongst the households in the 2017 NHTS sample.
Just over 4,000 (3%) of the households in the sample were missing a response to the household income question. For the purpose of this analysis they have been recoded with an income in the median income category, which has a midpoint of $62,500.
The second chart shows the relationship between AADVMT and household income and shows shows a clear postiive trend for both metropolitan and non-metropolitan households.
Figure 9: Distribution of Households in the 2017 NHTS by Household Income
Figure 10: 2017 NHTS AADVMT by Household Income
Household Size
Household size has a positive impact on the amount of travel a household makes, with larger households traveling more. The first chart shows the distribution of household size amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and household size and shows shows a clear postiive trend for both metropolitan and non-metropolitan households.
Figure 11: Distribution of Households in the 2017 NHTS by Household Size
Figure 12: 2017 NHTS AADVMT by Household Size
Household life cycle stage
The household’s life cycle stage, i.e., whether the household is a single person, a couple without children, a couple with children, or older “empty nesters” influences the amount of travel a household makes. Multi person household make more travel and adding children to the household causes a further moderate increase in the amount of travel. The first chart shows the distribution of life cycle stage amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and household life cycle stage for both metropolitan and non-metropolitan households.
Figure 13: Distribution of Households in the 2017 NHTS by Household Life Cycle Stage
Figure 14: 2017 NHTS AADVMT by Household Life Cycle Stage
Number of Workers
The number of workers in the household positively impacts the amount of travel a household makes. With each additional worker in the household, the household make more travel. The first chart shows the distribution of number of workers per households amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and number of workers per household for both metropolitan and non-metropolitan households.
Figure 15: Distribution of Households in the 2017 NHTS by Number of Workers in the Household
Figure 16: 2017 NHTS AADVMT by Number of Workers in the Household
Number of Drivers
The number of drivers in the household positively impacts the amount of travel a household makes. With each additional driver in the household, the household make more travel. The first chart shows the distribution of number of drivers per households amongst the households in the 2017 NHTS sample. The second charts shows the relationship between AADVMT and number of drivers per household for both metropolitan and non-metropolitan households.
Figure 17: Distribution of Households in the 2017 NHTS by Number of Drivers in the Household
Figure 18: 2017 NHTS AADVMT by Number of Drivers in the Household
Model Estimation
2017 NHTS AADVMT Model
The following table shows the AADVMT model estimated using the 2017 NHTS for Metro areas and Non-Metro area.
| Dependent variable: | ||
| AADVMT | ||
| NONMETRO | METRO | |
| (1) | (2) | |
| Drivers | 0.929*** (0.012) | 0.995*** (0.008) |
| HhSize | 0.105*** (0.010) | |
| Workers | 0.234*** (0.009) | 0.151*** (0.008) |
| CENSUS_RNE | -0.069*** (0.019) | -0.085*** (0.016) |
| CENSUS_RS | 0.095*** (0.014) | 0.012 (0.014) |
| CENSUS_RW | -0.212*** (0.017) | -0.119*** (0.015) |
| FwyLaneMiPC | 49.622*** (17.024) | |
| LogIncomeK | 0.351*** (0.007) | 0.190*** (0.006) |
| Age0to14 | -0.003 (0.012) | 0.090*** (0.009) |
| Age65Plus | -0.072*** (0.011) | -0.096*** (0.010) |
| log1p(VehPerDriver) | 4.105*** (0.025) | 4.262*** (0.021) |
| LifeCycleCouple w/o children | -0.033 (0.021) | -0.059*** (0.016) |
| LifeCycleEmpty Nester | -0.267*** (0.026) | -0.456*** (0.020) |
| LifeCycleSingle | -0.211*** (0.028) | -0.413*** (0.018) |
| D1B | -0.019*** (0.003) | -0.002*** (0.0002) |
| D2A_EPHHM | -0.298*** (0.029) | |
| D1B:D2A_EPHHM | 0.023*** (0.005) | |
| D2A_WRKEMP | -0.0002 (0.0002) | |
| D3bpo4 | -0.001*** (0.0001) | |
| TranRevMiPC:D4c | -0.056*** (0.004) | |
| Constant | -0.802*** (0.045) | -0.320*** (0.031) |
| Observations | 56,551 | 71,740 |
| R2 | 0.654 | 0.700 |
| Adjusted R2 | 0.654 | 0.700 |
| Residual Std. Error | 1.182 (df = 56534) | 1.384 (df = 71722) |
| F Statistic | 6,671.994*** (df = 16; 56534) | 9,861.924*** (df = 17; 71722) |
| Note: | p<0.1; p<0.05; p<0.01 | |
The following tables compare the AADVMT models estimated using the 2009 and 2017 NHTS.
This first comparison is between 2009 and 2017 AADVMT Models for Metro areas.
| VarName | NHTS2009 | NHTS2017 | Ratio |
|---|---|---|---|
| (Intercept) | -1.333 | -0.320 | 0.240 |
| Age0to14 | 0.107 | 0.090 | 0.840 |
| Age65Plus | -0.075 | -0.096 | 1.286 |
| CENSUS_RNE | -0.109 | -0.085 | 0.777 |
| CENSUS_RS | 0.051 | 0.012 | 0.227 |
| CENSUS_RW | -0.092 | -0.119 | 1.289 |
| D1B | -0.003 | -0.002 | 0.690 |
| D2A_WRKEMP | 0.000 | 0.000 | 0.914 |
| D3bpo4 | -0.001 | -0.001 | 0.892 |
| Drivers | 0.705 | 0.995 | 1.412 |
| FwyLaneMiPC | 101.341 | 49.622 | 0.490 |
| LifeCycleCouple w/o children | -0.036 | -0.059 | 1.611 |
| LifeCycleEmpty Nester | -0.256 | -0.456 | 1.778 |
| LifeCycleSingle | -0.234 | -0.413 | 1.767 |
| LogIncome | 0.268 | 0.000 | 0.000 |
| LogIncomeK | 0.000 | 0.190 | Inf |
| TranRevMiPC:D4c | -0.020 | -0.056 | 2.821 |
| Workers | 0.186 | 0.151 | 0.811 |
| log1p(VehPerDriver) | 1.794 | 4.262 | 2.376 |
This second comparison is between 2009 and 2017 AADVMT Models for Non-Metro areas.
| VarName | NHTS2009 | NHTS2017 | Ratio |
|---|---|---|---|
| (Intercept) | -1.416 | -0.802 | 0.566 |
| Age0to14 | 0.102 | -0.003 | -0.031 |
| Age65Plus | -0.077 | -0.072 | 0.932 |
| CENSUS_RNE | -0.112 | -0.069 | 0.617 |
| CENSUS_RS | 0.058 | 0.095 | 1.632 |
| CENSUS_RW | -0.176 | -0.212 | 1.206 |
| D1B | -0.008 | -0.019 | 2.325 |
| D1B:D2A_EPHHM | -0.027 | 0.023 | -0.867 |
| D2A_EPHHM | -0.084 | -0.298 | 3.524 |
| Drivers | 0.744 | 0.929 | 1.249 |
| HhSize | 0.017 | 0.105 | 6.231 |
| LifeCycleCouple w/o children | -0.013 | -0.033 | 2.421 |
| LifeCycleEmpty Nester | -0.208 | -0.267 | 1.281 |
| LifeCycleSingle | -0.216 | -0.211 | 0.975 |
| LogIncome | 0.288 | 0.000 | 0.000 |
| LogIncomeK | 0.000 | 0.351 | Inf |
| Workers | 0.177 | 0.234 | 1.322 |
| log1p(VehPerDriver) | 1.852 | 4.105 | 2.216 |
2017 NHTS AADVMT Model Adjustment
One aspect of the NHTS AADVMT data that is hard to capture, even with a power adjusted linear model, is the variability in the data due to unobserved household travel characteristics. Some households just travel more or less than other similar households with similar income, transportation access, and neighborhood characteristics.
In order to capture this dispersion, a random variable has been added to the model to simulate household to household variation. This is drawn from a left skewed normal distribution to allow for some households lower AADVMT, overall a slight increase in AADVMT due to under prediction of the mean AADVMT in the sample, and a longer tail of households with higher AADVMT.
ADD A CHART HERE TO EXPLAIN THIS
Model Prediction Testing
Households by Metro and Non-Metro
| Metro or Non-Metro | NHTS2017 AADVMT/HH | Model AADVMT/HH | Model Sim AADVMT/HH | Ratio Model/NHTS2017 | Ratio Model Sim/NHTS2017 |
|---|---|---|---|---|---|
| metro | 47.54 | 43.17 | 47.51 | 0.91 | 1.00 |
| non_metro | 61.04 | 56.58 | 62.27 | 0.93 | 1.02 |
Figure 19: Scatterplot of Model Prediction vs. 2017 NHTS Data
(#fig:scatter-aadvmt-model-prediction_rnd)Scatterplot of Model Prediction (Simulated) vs. 2017 NHTS Data
Figure 20: Households by DVMT Bins, Model Prediction vs. 2017 NHTS Data
Figure 21: Difference in Households in DVMT Bins (Model Prediction - 2017 NHTS Data)
Figure 22: Difference in Households in DVMT Bins (Model Prediction (Simulated) - 2017 NHTS Data)
Households by Income
| HH Income | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 5000 | 18.53 | 20.41 | 16.61 | 18.50 | 0.90 | 0.91 |
| 12500 | 21.29 | 29.89 | 21.57 | 28.53 | 1.01 | 0.95 |
| 19999 | 29.94 | 40.09 | 29.69 | 38.89 | 0.99 | 0.97 |
| 30000 | 37.60 | 48.02 | 36.64 | 47.35 | 0.97 | 0.99 |
| 42499 | 44.58 | 57.59 | 43.04 | 58.18 | 0.97 | 1.01 |
| 62500 | 48.06 | 65.62 | 47.83 | 67.94 | 1.00 | 1.04 |
| 87500 | 56.70 | 79.17 | 56.10 | 81.81 | 0.99 | 1.03 |
| 112500 | 64.35 | 88.30 | 65.36 | 87.59 | 1.02 | 0.99 |
| 137500 | 68.08 | 88.35 | 68.37 | 94.70 | 1.00 | 1.07 |
| 174999 | 68.91 | 91.53 | 71.45 | 96.63 | 1.04 | 1.06 |
| 249999 | 68.33 | 93.31 | 72.46 | 103.60 | 1.06 | 1.11 |
Figure 23: Households by Income Group, Model Prediction vs. 2017 NHTS Data
Figure 24: Scatterplot of Model Prediction vs. 2017 NHTS Data by Income
Households by Density
| D1B Group | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 0 | 53.35 | 61.42 | 53.50 | 62.65 | 1.00 | 1.02 |
| 10 | 44.74 | 48.65 | 44.93 | 49.55 | 1.00 | 1.02 |
| 20 | 39.94 | 48.37 | 38.60 | 43.21 | 0.97 | 0.89 |
| 30 | 33.59 | 36.09 | 31.26 | 59.09 | 0.93 | 1.64 |
| 40 | 30.18 | 13.51 | 29.69 | 27.42 | 0.98 | 2.03 |
| 50 | 23.74 | 0.00 | 24.71 | 0.00 | 1.04 | 0.00 |
| 60 | 22.28 | 0.00 | 24.81 | 0.00 | 1.11 | 0.00 |
| 70 | 20.13 | 0.00 | 19.50 | 0.00 | 0.97 | 0.00 |
| 80 | 20.74 | 0.00 | 20.20 | 0.00 | 0.97 | 0.00 |
| 90 | 17.75 | 0.00 | 19.67 | 0.00 | 1.11 | 0.00 |
| 100 | 14.92 | 0.00 | 13.87 | 0.00 | 0.93 | 0.00 |
| 200 | 9.43 | 0.00 | 7.62 | 0.00 | 0.81 | 0.00 |
Figure 25: Households by Density, Model Prediction vs. 2017 NHTS Data
Figure 26: Scatterplot of Model Prediction vs. 2017 NHTS Data by Density
Households by Household Size
| HH Size | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 1 | 22.38 | 28.57 | 22.61 | 29.52 | 1.01 | 1.03 |
| 2 | 47.44 | 59.50 | 46.18 | 59.05 | 0.97 | 0.99 |
| 3 | 61.62 | 78.82 | 60.94 | 78.35 | 0.99 | 0.99 |
| 4 | 70.45 | 84.68 | 71.89 | 88.19 | 1.02 | 1.04 |
| 5 | 75.39 | 85.51 | 78.77 | 92.62 | 1.04 | 1.08 |
| 6 | 81.19 | 87.61 | 83.22 | 97.84 | 1.02 | 1.12 |
Figure 27: Households by Size, Model Prediction vs. 2017 NHTS Data
Figure 28: Scatterplot of Model Prediction vs. 2017 NHTS Data by Household Size
Households by Number of Workers
| Workers | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 0 | 23.26 | 34.24 | 23.66 | 34.25 | 1.02 | 1.00 |
| 1 | 41.30 | 56.57 | 40.18 | 56.16 | 0.97 | 0.99 |
| 2 | 66.00 | 83.73 | 64.95 | 84.77 | 0.98 | 1.01 |
| 3 | 90.39 | 111.44 | 96.76 | 123.96 | 1.07 | 1.11 |
| 4 | 106.01 | 130.78 | 125.68 | 164.80 | 1.19 | 1.26 |
| 5 | 129.07 | 141.05 | 163.93 | 207.31 | 1.27 | 1.47 |
Figure 29: Households by Workers, Model Prediction vs. 2017 NHTS Data
Figure 30: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Workers
Households by Number of Drivers
| Drivers | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 0 | 0.16 | 0.79 | 0.01 | 0.10 | 0.09 | 0.13 |
| 1 | 28.85 | 33.85 | 29.63 | 35.57 | 1.03 | 1.05 |
| 2 | 58.82 | 69.71 | 56.62 | 68.12 | 0.96 | 0.98 |
| 3 | 84.46 | 97.00 | 85.70 | 101.77 | 1.01 | 1.05 |
| 4 | 108.70 | 120.81 | 122.02 | 147.77 | 1.12 | 1.22 |
| 5 | 132.09 | 135.04 | 164.48 | 179.12 | 1.25 | 1.33 |
Figure 31: Households by Drivers, Model Prediction vs. 2017 NHTS Data
Figure 32: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Drivers
Households by Number of Vehicles
| Vehicles | Metro Obs AADVMT/HH | Non-Metro Obs AADVMT/HH | Metro Model AADVMT/HH | Non-Metro Model AADVMT/HH | Metro Ratio Model/Obs | Non-Metro Ratio Model/Obs |
|---|---|---|---|---|---|---|
| 0 | 0.00 | 0.00 | 0.89 | 1.60 | Inf | Inf |
| 1 | 29.24 | 30.42 | 27.81 | 28.45 | 0.95 | 0.94 |
| 2 | 61.01 | 63.28 | 59.47 | 63.53 | 0.97 | 1.00 |
| 3 | 84.56 | 90.47 | 87.69 | 93.08 | 1.04 | 1.03 |
| 4 | 109.21 | 114.26 | 118.59 | 126.65 | 1.09 | 1.11 |
| 5 | 127.92 | 136.72 | 143.26 | 147.25 | 1.12 | 1.08 |
| 6 | 144.54 | 139.28 | 154.00 | 157.15 | 1.07 | 1.13 |
| 7 | 129.45 | 129.46 | 137.40 | 124.78 | 1.06 | 0.96 |
| 8 | 176.98 | 154.07 | 170.35 | 167.93 | 0.96 | 1.09 |
| 9 | 106.11 | 206.69 | 170.72 | 170.19 | 1.61 | 0.82 |
| 10 | 10.89 | 124.34 | 65.21 | 115.68 | 5.99 | 0.93 |
| 11 | 151.46 | 209.79 | 112.26 | 129.56 | 0.74 | 0.62 |
| 12 | 111.86 | 210.32 | 83.72 | 128.65 | 0.75 | 0.61 |
Figure 33: Households by Drivers, Model Prediction vs. 2017 NHTS Data
Figure 34: Scatterplot of Model Prediction vs. 2017 NHTS Data by Number of Vehicles
Benchmarking Against 2009 Multimodal Model
The following tables and chart compare the performance of the 2009 and 2017 AADVMT Models by applying the 2009 model to the 2017 NHTS households.
| Metro or Non-Metro | Num HH | NHTS2017 AADVMT/HH | Model (2017) AADVMT/HH | Model (2009) AADVMT/HH | Ratio Model (2017)/NHTS2017 | Ratio Model (2009)/NHTS2017 |
|---|---|---|---|---|---|---|
| metro | 71740 | 47.54 | 47.51 | 42.37 | 1.00 | 0.89 |
| non_metro | 56538 | 61.04 | 62.27 | 53.65 | 1.02 | 0.88 |
Figure 35: Scatterplot of 2009 Model Prediction vs. 2017 NHTS Data
Figure 36: Scatterplot of 2009 Model Prediction vs. 2017 Model Prediction